Abstract
The chapter will propose a novel approach that combines the traditional machine learning approach in churn management and customer satisfaction evaluation, which unite traditional machine learning approach and expert-based approach, which leans on event-based management. The core of the proposed framework is hybrid fuzzy expert system, which can contain a variety of data mining predictive models responsible for some specific areas as additions to traditional rule blocks. It can also include social network analysis metrics based on linguistic variables and incorporated within rule blocks. The chapter will introduce how revealed patterns can be applied for continual portfolio management improvement. The proposed solution unites advanced analytical techniques with the decision-making process within a holistic self-learning framework.
TopIntroduction
Big data environment gives new opportunities in analytics. From one hand it demands new way of thinking in combination of unstructured data sources. Decision support system which could unite most important factors and find sensitiveness between recognized factors could be useful decision support tool for portfolio management. Tool which can predict, with certain level of probability potential values of observed outputs, regarding recognized inputs, or simulated inputs can be valuable tool for decision.
Unstructured data sources bring new approaches and opportunities in better understanding business areas like early warning systems. Main advantage challenge even we are talking about big data environment to recognize rare events in such environment which statistically does not have a major influence on target attributes, but when it happened, they have huge impact on target attribute like churn.
Example for that can be complaints on service delivery. In general, we do not expect statistically huge number of complaints in regular business, except in extreme situation. Among other information, share of complaints within data as cases are rare. Traditional machine learning algorithms potentially during machine learning process will not recognize it as important regarding low occurrence ratio. Generally speaking, such events can be very important for detection situations like churn. Here we can talk about paradox in big data environment where significant attributes and values in big data environment cannot be adequately incorporated into machine learning models because of rare value occurrence.
Value of the existing data, dislocated within different transactional systems could be increased by integration into data lake. It still does not mean that company does not have limited information about some problem space. Other problem in relation with traditional analytical approach is often avoidance of unstructured data source usage for business modeling purposes, even unstructured data exists within systems like call centers data or similar sources.
Modern decade fights with problem how to interact and efficiently use large available data collections.
Also in our case we are talking about adequate recognition and usage of important, but rare events.
Such conditions demands combination industry and expert knowledge but are easy to use. Described as Early Warning Systems (EWS) those models are packed with state-of-the-art knowledge and KPI’s which helps business people to deal with numerous influences, large data sets and market trends. Early Warning Systems are also in close connection with risk assessment and modeling. Risk management methods often precede development phase and are used as starting point for early warning systems development.
By dealing with knowledge, term intelligence and, furthermore, business intelligence needs to be introduced first. Intelligence concerns the awareness and knowledge of the external business environment. The definition that is used here is that business intelligence is a systematized and continuous approach to focus, collect, analyze, communicate, and use information about customers, competitors, distributors, technology, political issues, macroeconomic issues, and political issues in order to increase the competitiveness of the organization (Hedin, Kovero 2006).
Key Terms in this Chapter
EWS: Early Warning Systems.
Social Network Bridge: Individual, whose weak ties fill a structural hole, providing the only link between two individuals or clusters.
Betweenness Centrality: Measure of a node's centrality in a network; it is equal to the number of shortest paths from all vertices to all others that pass through that node.
Data Mining: Discovering hidden useful knowledge in large amount of data (databases).
Fuzzy Expert System: Expert system based on fuzzy logic.
Social Network Centrality: Refers to a group of metrics that aim to quantify the “importance” or “influence” (in a variety of senses) of a particular node (or group) within a network.
SNA: Social network analysis.
Fuzzy Logic: Logic which presumes possible membership to more than one category with degree of membership, and which is opposite to (exact) crisp logic.